[💡SUG] Support for implicit feedback context aware recommender

[samruddha-kf]

Is your feature request related to a problem? Please describe. This is more of a question than request. I have a implicit feedback dataset, where interaction context are quite important. For example a season or time of day, holiday, weekend affects the occurrence of an interaction. After experimenting quite a bit, my understanding is, library supports having context fields for user and/or item but one has to create an custom model to add context fields corresponding to interaction. So I also tried that, but I further learned that, label or rating field is still a requirement, since it is not possible to have interaction context for negative sample. Please let me know if I am wrong and there is a way I can use the implicit feedback dataset and interaction contexts in training a recommender (any of general, context or sequential (i do have timestamp data)).

Describe the solution you'd like As described above, I am not expecting a solution. But does it make sense (theoretically) to generate the all possible combinations of interaction context for unobserved interactions. If this makes sense, then augmenting dataset module could be added.

[TayTroye]

@samruddha-kf Hello,thanks for your attention to RecBole.Now the datasets used in the context-aware models are explicit.The use of these contexts should be model-specific.Subsequently, RecBle will consider adding new moduels.